Deterioration detection and safety assessment for reinforced concrete structures caused by rebar corrosion in coastal areas based on YOLOv7-AM

Rebar corrosion threatens coastal reinforced concrete (RC) structures. Traditional inspection methods are laborious, while deep learning often focuses on load-induced damage, neglecting environmental deterioration. This study bridges the gap between surface damage detection and durability assessment. First, relationships between rebar mass loss rate and component bearing capacity were established based on 398 RC beam and column tests. Second, a durability surface deterioration dataset (DSD) with 5368 images of normal cracks, corrosion cracks, and spalling was developed. Third, an improved YOLOv7-AM network integrating Convolutional Block Attention Module (CBAM) was proposed, achieving mean average precisions of 90.02 % and 91.21 % at 640 × 640 and 800 × 800 pixel inputs, outperforming mainstream models in multi-class damage recognition. Finally, a novel system combining damage identification with durability assessment was constructed, enabling automated safety evaluation. This study provides an efficient, accurate solution for coastal RC structure maintenance, significantly advancing corrosion-induced deterioration assessment.